brand-kit — Stage 1 + 2 research skill (PMM spine)

Compiles visual identity into a structured brand kit. Output format: DESIGN.md (YAML token frontmatter + 8 prose sections). The YAML tokens are the machine-authoritative source — downstream skills that produce visual output (landing-page wireframes, ad creative briefs, slide decks) consume the tokens directly.

When to use

• When the team refreshes visual identity (new color palette, new typography system)
• Before building any landing page, ad creative, or designed artifact — they need the brand kit as input
• Quarterly refresh

When NOT to use

• For voice / tone (use /tov-guidelines — voice + visual are separate but both live in marketing/brand/)
• For positioning or messaging (use /positioning or /product-messaging)
• For one-off design checks (write a custom gate skill that reads brand-kit.md tokens)

How it works

1. Inputs: screenshots of your product / website (primary, higher fidelity) OR website URL (fallback, lower fidelity).
2. Reads: visual content via Claude's vision (screenshots) or Exa MCP web fetch (URL).
3. Extracts: exact colors (hex codes), typography stack (font family + sizes + weights + line heights + letter spacing), spacing scale, corner radii, component patterns (button states, card styles, input styles, etc.).
4. Produces a DESIGN.md-format brand kit:

• YAML token frontmatter (machine-authoritative): colors, typography, rounded, spacing, components — token references use {path.to.token} syntax
• 8 ordered prose sections: Overview, Colors, Typography, Layout, Elevation & Depth, Shapes, Components, Do's and Don'ts

1. Validates against the design-production doctrine: ≤2 brand colors active per screen, ≤2 font weights per screen, ≤10% accent coverage, no banned patterns (gradient text, side-stripe borders, hero-metric template, icon-tile-above-heading template, cardocalypse, glassmorphism).
2. Writes to marketing/brand/brand-kit.md (overwrites prior canonical; git history preserves prior versions).

Invoke

/brand-kit screenshots: [paths to product screenshots]

Or URL-fallback:

/brand-kit https://yourdomain.com — extract visual identity

Example output

See marketing/brand/brand-kit.md for the PulseAnalytics seed. Notice the YAML frontmatter is the source of truth; the prose body explains intent. Token references use {path.to.token} syntax — every component reference resolves to a defined token. Do's and Don'ts encode quantitative limits (≤2 font weights per screen, ≤10% accent coverage).

Dependencies

• Reads from: source visual content (screenshots or URL — provided as input)
• Reads via Exa MCP (optional): website if URL-fallback path is used
• Writes to: marketing/brand/brand-kit.md (canonical)
• Downstream readers: /landing-page-wireframe, /landing-page-copy, /ad-creative-brief, /sales-deck, /linkedin-carousels, any visual-output skill

Customization

Add custom token categories if your brand requires them (e.g., motion tokens for animation timing, iconography tokens for icon style). The YAML frontmatter accepts arbitrary top-level keys; downstream skills consume whatever exists.

Update the Do's and Don'ts when you spot a recurring violation in produced artifacts — add a rule, downstream skills enforce it on next run.

Where this fits in the Example 1 chain

Week 1: /brand-kit (parallel with positioning + voice research) → visual identity locked
Week 2: /landing-page-wireframe + /landing-page-copy read brand kit + messaging
Week 3+: every visual-output skill reads brand kit

Refresh cadence

Quarterly. Trigger sooner on visual identity refresh, new product launch with distinct visual treatment, or major brand update.

Design production doctrine

See .claude/rules/design-production.md (in the Genesys workspace; not duplicated here) for the integration contract — how YAML tokens flow into CSS variables (web), Figma variables (design), and slide / image templates (non-web).